src/main.c

   1 #ifdef HAVE_CONFIG_H
   2 #include <config.h>
   3 #endif
   4 #include <stdio.h>
   5 #include <stdlib.h>
   6 #include <unistd.h>
   7 #include <signal.h>
   8 #include <syslog.h>
   9 #include <pthread.h>
  10 #include <string.h>
  11 #include <errno.h>
  12 #include <sys/types.h>
  13 #include <sys/wait.h>
  14 #include <sys/time.h>
  15 #include <sys/resource.h>
  16 #include <sys/mman.h>
  17 #include <sys/prctl.h>
  18
  19 #include "common.h"
  20 #include "log.h"
  21 #include "daemon.h"
  22 #include "web_server.h"
  23 #include "popen.h"
  24 #include "appconfig.h"
  25 #include "web_client.h"
  26 #include "rrd.h"
  27 #include "rrd2json.h"
  28
  29 #include "unit_test.h"
  30
  31 #include "plugins_d.h"
  32 #include "plugin_idlejitter.h"
  33 #include "plugin_tc.h"
  34 #include "plugin_checks.h"
  35 #include "plugin_proc.h"
  36 #include "plugin_nfacct.h"
  37
  38 #include "main.h"
  39
  40 int netdata_exit = 0;
  41
  42 void netdata_cleanup_and_exit(int ret)
  43 {
  44         netdata_exit = 1;
  45         rrdset_save_all();
  46         // kill_childs();
  47         unlink("/var/run/netdata.pid");
  48         info("NetData exiting. Bye bye...");
  49         exit(ret);
  50 }
  51
  52 struct netdata_static_thread {
  53         char *name;
  54
  55         char *config_section;
  56         char *config_name;
  57
  58         int enabled;
  59
  60         pthread_t *thread;
  61
  62         void (*init_routine) (void);
  63         void *(*start_routine) (void *);
  64 };
  65
  66 struct netdata_static_thread static_threads[] = {
  67         {"tc",                  "plugins",      "tc",                   1, NULL, NULL,  tc_main},
  68         {"idlejitter",  "plugins",      "idlejitter",   1, NULL, NULL,  cpuidlejitter_main},
  69         {"proc",                "plugins",      "proc",                 1, NULL, NULL,  proc_main},
  70
  71 #ifdef INTERNAL_PLUGIN_NFACCT
  72         // nfacct requires root access
  73         // so, we build it as an external plugin with setuid to root
  74         {"nfacct",              "plugins",      "nfacct",               1, NULL, NULL,  nfacct_main},
  75 #endif
  76
  77         {"plugins.d",   NULL,           NULL,                   1, NULL, NULL,  pluginsd_main},
  78         {"check",               "plugins",      "checks",               0, NULL, NULL,  checks_main},
  79         {"web",                 NULL,           NULL,                   1, NULL, NULL,  socket_listen_main},
  80         {NULL,                  NULL,           NULL,                   0, NULL, NULL,  NULL}
  81 };
  82
  83 int killpid(pid_t pid, int sig)
  84 {
  85         int ret = -1;
  86         debug(D_EXIT, "Request to kill pid %d", pid);
  87
  88         errno = 0;
  89         if(kill(pid, 0) == -1) {
  90                 switch(errno) {
  91                         case ESRCH:
  92                                 error("Request to kill pid %d, but it is not running.", pid);
  93                                 break;
  94
  95                         case EPERM:
  96                                 error("Request to kill pid %d, but I do not have enough permissions.", pid);
  97                                 break;
  98
  99                         default:
 100                                 error("Request to kill pid %d, but I received an error.", pid);
 101                                 break;
 102                 }
 103         }
 104         else {
 105                 errno = 0;
 106
 107                 void (*old)(int);
 108                 old = signal(sig, SIG_IGN);
 109                 if(old == SIG_ERR) {
 110                         error("Cannot overwrite signal handler for signal %d", sig);
 111                         old = sig_handler;
 112                 }
 113
 114                 ret = kill(pid, sig);
 115
 116                 if(signal(sig, old) == SIG_ERR)
 117                         error("Cannot restore signal handler for signal %d", sig);
 118
 119                 if(ret == -1) {
 120                         switch(errno) {
 121                                 case ESRCH:
 122                                         error("Cannot kill pid %d, but it is not running.", pid);
 123                                         break;
 124
 125                                 case EPERM:
 126                                         error("Cannot kill pid %d, but I do not have enough permissions.", pid);
 127                                         break;
 128
 129                                 default:
 130                                         error("Cannot kill pid %d, but I received an error.", pid);
 131                                         break;
 132                         }
 133                 }
 134         }
 135
 136         return ret;
 137 }
 138
 139 void kill_childs()
 140 {
 141         siginfo_t info;
 142
 143         struct web_client *w;
 144         for(w = web_clients; w ; w = w->next) {
 145                 debug(D_EXIT, "Stopping web client %s", w->client_ip);
 146                 pthread_cancel(w->thread);
 147                 pthread_join(w->thread, NULL);
 148         }
 149
 150         int i;
 151         for (i = 0; static_threads[i].name != NULL ; i++) {
 152                 if(static_threads[i].thread) {
 153                         debug(D_EXIT, "Stopping %s thread", static_threads[i].name);
 154                         pthread_cancel(*static_threads[i].thread);
 155                         pthread_join(*static_threads[i].thread, NULL);
 156                         static_threads[i].thread = NULL;
 157                 }
 158         }
 159
 160         if(tc_child_pid) {
 161                 info("Killing tc-qos-helper procees");
 162                 if(killpid(tc_child_pid, SIGTERM) != -1)
 163                         waitid(P_PID, (id_t) tc_child_pid, &info, WEXITED);
 164         }
 165         tc_child_pid = 0;
 166
 167         struct plugind *cd;
 168         for(cd = pluginsd_root ; cd ; cd = cd->next) {
 169                 debug(D_EXIT, "Stopping %s plugin thread", cd->id);
 170                 pthread_cancel(cd->thread);
 171                 pthread_join(cd->thread, NULL);
 172
 173                 if(cd->pid && !cd->obsolete) {
 174                         debug(D_EXIT, "killing %s plugin process", cd->id);
 175                         if(killpid(cd->pid, SIGTERM) != -1)
 176                                 waitid(P_PID, (id_t) cd->pid, &info, WEXITED);
 177                 }
 178         }
 179
 180         // if, for any reason there is any child exited
 181         // catch it here
 182         waitid(P_PID, 0, &info, WEXITED|WNOHANG);
 183
 184         debug(D_EXIT, "All threads/childs stopped.");
 185 }
 186
 187
 188 int main(int argc, char **argv)
 189 {
 190         int i;
 191         int config_loaded = 0;
 192         int dont_fork = 0;
 193
 194         // global initialization
 195         get_HZ();
 196
 197         // set the name for logging
 198         program_name = "netdata";
 199
 200         // parse  the arguments
 201         for(i = 1; i < argc ; i++) {
 202                 if(strcmp(argv[i], "-c") == 0 && (i+1) < argc) {
 203                         if(load_config(argv[i+1], 1) != 1) {
 204                                 error("Cannot load configuration file %s.", argv[i+1]);
 205                                 exit(1);
 206                         }
 207                         else {
 208                                 debug(D_OPTIONS, "Configuration loaded from %s.", argv[i+1]);
 209                                 config_loaded = 1;
 210                         }
 211                         i++;
 212                 }
 213                 else if(strcmp(argv[i], "-df") == 0 && (i+1) < argc) { config_set("global", "debug flags",  argv[i+1]); debug_flags = strtoull(argv[i+1], NULL, 0); i++; }
 214                 else if(strcmp(argv[i], "-p")  == 0 && (i+1) < argc) { config_set("global", "port",         argv[i+1]); i++; }
 215                 else if(strcmp(argv[i], "-u")  == 0 && (i+1) < argc) { config_set("global", "run as user",  argv[i+1]); i++; }
 216                 else if(strcmp(argv[i], "-l")  == 0 && (i+1) < argc) { config_set("global", "history",      argv[i+1]); i++; }
 217                 else if(strcmp(argv[i], "-t")  == 0 && (i+1) < argc) { config_set("global", "update every", argv[i+1]); i++; }
 218                 else if(strcmp(argv[i], "-ch") == 0 && (i+1) < argc) { config_set("global", "host access prefix", argv[i+1]); i++; }
 219                 else if(strcmp(argv[i], "-nodaemon") == 0 || strcmp(argv[i], "-nd") == 0) dont_fork = 1;
 220                 else if(strcmp(argv[i], "--unittest")  == 0) {
 221                         rrd_update_every = 1;
 222                         if(run_all_mockup_tests()) exit(1);
 223                         if(unit_test_storage()) exit(1);
 224                         fprintf(stderr, "\n\nALL TESTS PASSED\n\n");
 225                         exit(0);
 226                 }
 227                 else {
 228                         fprintf(stderr, "Cannot understand option '%s'.\n", argv[i]);
 229                         fprintf(stderr, "\nUSAGE: %s [-d] [-l LINES_TO_SAVE] [-u UPDATE_TIMER] [-p LISTEN_PORT] [-df debug flags].\n\n", argv[0]);
 230                         fprintf(stderr, "  -c CONFIG FILE the configuration file to load. Default: %s.\n", CONFIG_DIR "/" CONFIG_FILENAME);
 231                         fprintf(stderr, "  -l LINES_TO_SAVE can be from 5 to %d lines in JSON data. Default: %d.\n", RRD_HISTORY_ENTRIES_MAX, RRD_DEFAULT_HISTORY_ENTRIES);
 232                         fprintf(stderr, "  -t UPDATE_TIMER can be from 1 to %d seconds. Default: %d.\n", UPDATE_EVERY_MAX, UPDATE_EVERY);
 233                         fprintf(stderr, "  -p LISTEN_PORT can be from 1 to %d. Default: %d.\n", 65535, LISTEN_PORT);
 234                         fprintf(stderr, "  -u USERNAME can be any system username to run as. Default: none.\n");
 235                         fprintf(stderr, "  -ch path to access host /proc and /sys when running in a container. Default: empty.\n");
 236                         fprintf(stderr, "  -nd or -nodeamon to disable forking in the background. Default: unset.\n");
 237                         fprintf(stderr, "  -df FLAGS debug options. Default: 0x%08llx.\n", debug_flags);
 238                         exit(1);
 239                 }
 240         }
 241
 242         if(!config_loaded) load_config(NULL, 0);
 243
 244         // prepare configuration environment variables for the plugins
 245         setenv("NETDATA_CONFIG_DIR" , config_get("global", "config directory"   , CONFIG_DIR) , 1);
 246         setenv("NETDATA_PLUGINS_DIR", config_get("global", "plugins directory"  , PLUGINS_DIR), 1);
 247         setenv("NETDATA_WEB_DIR"    , config_get("global", "web files directory", WEB_DIR)    , 1);
 248         setenv("NETDATA_CACHE_DIR"  , config_get("global", "cache directory"    , CACHE_DIR)  , 1);
 249         setenv("NETDATA_LOG_DIR"    , config_get("global", "log directory"      , LOG_DIR)    , 1);
 250         setenv("NETDATA_HOST_PREFIX", config_get("global", "host access prefix" , "")         , 1);
 251
 252         // avoid extended to stat(/etc/localtime)
 253         // http://stackoverflow.com/questions/4554271/how-to-avoid-excessive-stat-etc-localtime-calls-in-strftime-on-linux
 254         setenv("TZ", ":/etc/localtime", 0);
 255
 256         // cd to /tmp to avoid any plugins writing files at random places
 257         if(chdir("/tmp")) error("netdata: ERROR: Cannot cd to /tmp");
 258
 259         char *input_log_file = NULL;
 260         char *output_log_file = NULL;
 261         char *error_log_file = NULL;
 262         char *access_log_file = NULL;
 263         char *user = NULL;
 264         {
 265                 char buffer[1024];
 266
 267                 // --------------------------------------------------------------------
 268
 269                 sprintf(buffer, "0x%08llx", 0ULL);
 270                 char *flags = config_get("global", "debug flags", buffer);
 271                 setenv("NETDATA_DEBUG_FLAGS", flags, 1);
 272
 273                 debug_flags = strtoull(flags, NULL, 0);
 274                 debug(D_OPTIONS, "Debug flags set to '0x%8llx'.", debug_flags);
 275
 276                 if(debug_flags != 0) {
 277                         struct rlimit rl = { RLIM_INFINITY, RLIM_INFINITY };
 278                         if(setrlimit(RLIMIT_CORE, &rl) != 0)
 279                                 info("Cannot request unlimited core dumps for debugging... Proceeding anyway...");
 280                         prctl(PR_SET_DUMPABLE, 1, 0, 0, 0);
 281                 }
 282
 283                 // --------------------------------------------------------------------
 284
 285 #ifdef MADV_MERGEABLE
 286                 enable_ksm = config_get_boolean("global", "memory deduplication (ksm)", enable_ksm);
 287 #else
 288 #warning "Kernel memory deduplication (KSM) is not available"
 289 #endif
 290
 291                 // --------------------------------------------------------------------
 292
 293
 294                 global_host_prefix = config_get("global", "host access prefix", "");
 295                 setenv("NETDATA_HOST_PREFIX", global_host_prefix, 1);
 296
 297                 // --------------------------------------------------------------------
 298
 299                 output_log_file = config_get("global", "debug log", LOG_DIR "/debug.log");
 300                 if(strcmp(output_log_file, "syslog") == 0) {
 301                         output_log_syslog = 1;
 302                         output_log_file = NULL;
 303                 }
 304                 else if(strcmp(output_log_file, "none") == 0) {
 305                         output_log_syslog = 0;
 306                         output_log_file = NULL;
 307                 }
 308                 else output_log_syslog = 0;
 309
 310                 // --------------------------------------------------------------------
 311
 312                 error_log_file = config_get("global", "error log", LOG_DIR "/error.log");
 313                 if(strcmp(error_log_file, "syslog") == 0) {
 314                         error_log_syslog = 1;
 315                         error_log_file = NULL;
 316                 }
 317                 else if(strcmp(error_log_file, "none") == 0) {
 318                         error_log_syslog = 0;
 319                         error_log_file = NULL;
 320                         // optimization - do not even generate debug log entries
 321                 }
 322                 else error_log_syslog = 0;
 323
 324                 // --------------------------------------------------------------------
 325
 326                 access_log_file = config_get("global", "access log", LOG_DIR "/access.log");
 327                 if(strcmp(access_log_file, "syslog") == 0) {
 328                         access_log_syslog = 1;
 329                         access_log_file = NULL;
 330                 }
 331                 else if(strcmp(access_log_file, "none") == 0) {
 332                         access_log_syslog = 0;
 333                         access_log_file = NULL;
 334                 }
 335                 else access_log_syslog = 0;
 336
 337                 // --------------------------------------------------------------------
 338
 339                 rrd_memory_mode = rrd_memory_mode_id(config_get("global", "memory mode", rrd_memory_mode_name(rrd_memory_mode)));
 340
 341                 // --------------------------------------------------------------------
 342
 343                 if(gethostname(buffer, HOSTNAME_MAX) == -1)
 344                         error("WARNING: Cannot get machine hostname.");
 345                 hostname = config_get("global", "hostname", buffer);
 346                 debug(D_OPTIONS, "hostname set to '%s'", hostname);
 347
 348                 // --------------------------------------------------------------------
 349
 350                 rrd_default_history_entries = (int) config_get_number("global", "history", RRD_DEFAULT_HISTORY_ENTRIES);
 351                 if(rrd_default_history_entries < 5 || rrd_default_history_entries > RRD_HISTORY_ENTRIES_MAX) {
 352                         fprintf(stderr, "Invalid save lines %d given. Defaulting to %d.\n", rrd_default_history_entries, RRD_DEFAULT_HISTORY_ENTRIES);
 353                         rrd_default_history_entries = RRD_DEFAULT_HISTORY_ENTRIES;
 354                 }
 355                 else {
 356                         debug(D_OPTIONS, "save lines set to %d.", rrd_default_history_entries);
 357                 }
 358
 359                 // --------------------------------------------------------------------
 360
 361                 rrd_update_every = (int) config_get_number("global", "update every", UPDATE_EVERY);
 362                 if(rrd_update_every < 1 || rrd_update_every > 600) {
 363                         fprintf(stderr, "Invalid update timer %d given. Defaulting to %d.\n", rrd_update_every, UPDATE_EVERY_MAX);
 364                         rrd_update_every = UPDATE_EVERY;
 365                 }
 366                 else debug(D_OPTIONS, "update timer set to %d.", rrd_update_every);
 367
 368                 // let the plugins know the min update_every
 369                 {
 370                         char buf[50];
 371                         snprintf(buf, 50, "%d", rrd_update_every);
 372                         setenv("NETDATA_UPDATE_EVERY", buf, 1);
 373                 }
 374
 375                 // --------------------------------------------------------------------
 376
 377                 for (i = 0; static_threads[i].name != NULL ; i++) {
 378                         struct netdata_static_thread *st = &static_threads[i];
 379
 380                         if(st->config_name) st->enabled = config_get_boolean(st->config_section, st->config_name, st->enabled);
 381                         if(st->enabled && st->init_routine) st->init_routine();
 382                 }
 383
 384                 // --------------------------------------------------------------------
 385
 386                 prepare_rundir();
 387                 user = config_get("global", "run as user"    , (getuid() == 0)?NETDATA_USER:"");
 388                 web_files_uid();
 389
 390                 // --------------------------------------------------------------------
 391
 392                 listen_backlog = (int) config_get_number("global", "http port listen backlog", LISTEN_BACKLOG);
 393
 394                 listen_port = (int) config_get_number("global", "port", LISTEN_PORT);
 395                 if(listen_port < 1 || listen_port > 65535) {
 396                         fprintf(stderr, "Invalid listen port %d given. Defaulting to %d.\n", listen_port, LISTEN_PORT);
 397                         listen_port = LISTEN_PORT;
 398                 }
 399                 else debug(D_OPTIONS, "Listen port set to %d.", listen_port);
 400
 401                 int ip = 0;
 402                 char *ipv = config_get("global", "ip version", "any");
 403                 if(!strcmp(ipv, "any") || !strcmp(ipv, "both") || !strcmp(ipv, "all")) ip = 0;
 404                 else if(!strcmp(ipv, "ipv4") || !strcmp(ipv, "IPV4") || !strcmp(ipv, "IPv4") || !strcmp(ipv, "4")) ip = 4;
 405                 else if(!strcmp(ipv, "ipv6") || !strcmp(ipv, "IPV6") || !strcmp(ipv, "IPv6") || !strcmp(ipv, "6")) ip = 6;
 406                 else fprintf(stderr, "Cannot understand ip version '%s'. Assumming 'any'.", ipv);
 407
 408                 if(ip == 0 || ip == 6) listen_fd = create_listen_socket6(listen_port, listen_backlog);
 409                 if(listen_fd < 0) {
 410                         listen_fd = create_listen_socket4(listen_port, listen_backlog);
 411                         if(listen_fd >= 0 && ip != 4) fprintf(stderr, "Managed to open an IPv4 socket on port %d.", listen_port);
 412                 }
 413
 414                 if(listen_fd < 0) fatal("Cannot listen socket.");
 415         }
 416
 417         // never become a problem
 418         if(nice(20) == -1) error("Cannot lower my CPU priority.");
 419
 420         if(become_daemon(dont_fork, 0, user, input_log_file, output_log_file, error_log_file, access_log_file, &access_fd, &stdaccess) == -1) {
 421                 fatal("Cannot demonize myself.");
 422                 exit(1);
 423         }
 424
 425         if(debug_flags != 0) {
 426                 struct rlimit rl = { RLIM_INFINITY, RLIM_INFINITY };
 427                 if(setrlimit(RLIMIT_CORE, &rl) != 0)
 428                         info("Cannot request unlimited core dumps for debugging... Proceeding anyway...");
 429
 430                 prctl(PR_SET_DUMPABLE, 1, 0, 0, 0);
 431         }
 432
 433         if(output_log_syslog || error_log_syslog || access_log_syslog)
 434                 openlog("netdata", LOG_PID, LOG_DAEMON);
 435
 436         info("NetData started on pid %d", getpid());
 437
 438
 439         // catch all signals
 440         for (i = 1 ; i < 65 ;i++) {
 441                 switch(i) {
 442                         case SIGKILL: // not catchable
 443                         case SIGSTOP: // not catchable
 444                                 break;
 445
 446                         case SIGSEGV:
 447                         case SIGFPE:
 448                         case SIGCHLD:
 449                                 signal(i, SIG_DFL);
 450                                 break;
 451
 452                         default:
 453                                 signal(i,  sig_handler);
 454                                 break;
 455                 }
 456         }
 457
 458         for (i = 0; static_threads[i].name != NULL ; i++) {
 459                 struct netdata_static_thread *st = &static_threads[i];
 460
 461                 if(st->enabled) {
 462                         st->thread = malloc(sizeof(pthread_t));
 463                         if(!st->thread)
 464                                 fatal("Cannot allocate pthread_t memory");
 465
 466                         info("Starting thread %s.", st->name);
 467
 468                         if(pthread_create(st->thread, NULL, st->start_routine, NULL))
 469                                 error("failed to create new thread for %s.", st->name);
 470
 471                         else if(pthread_detach(*st->thread))
 472                                 error("Cannot request detach of newly created %s thread.", st->name);
 473                 }
 474                 else info("Not starting thread %s.", st->name);
 475         }
 476
 477         // for future use - the main thread
 478         while(1) {
 479                 if(netdata_exit != 0) {
 480                         netdata_exit++;
 481
 482                         if(netdata_exit > 5) {
 483                                 netdata_cleanup_and_exit(0);
 484                                 exit(0);
 485                         }
 486                 }
 487                 sleep(2);
 488         }
 489
 490         exit(0);
 491 }